Linux kernel mirror (for testing) git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel os linux
1
fork

Configure Feed

Select the types of activity you want to include in your feed.

pwm: stm32: Improve precision of calculation in .apply()

While mathematically it's ok to calculate the number of cyles for the
duty cycle as:

duty_cycles = period_cycles * duty_ns / period_ns

this doesn't always give the right result when doing integer math. This
is best demonstrated using an example: With the input clock running at
208877930 Hz a request for duty_cycle = 383 ns and period = 49996 ns
results in

period_cycles = clkrate * period_ns / NSEC_PER_SEC = 10443.06098828

Now calculating duty_cycles with the above formula gives:

duty_cycles = 10443.06098828 * 383 / 49996 = 80.00024719

However with period_cycle truncated to an integer results in:

duty_cycles = 10443 * 383 / 49996 = 79.99977998239859

So while a value of (a little more than) 80 would be the right result,
only 79 is used here. The problem here is that 14443 is a rounded result
that should better not be used to do further math. So to fix that use
the exact formular similar to how period_cycles is calculated.

Link: https://lore.kernel.org/r/7628ecd8a7538aa5a7397f0fc4199a077168e8a6.1710711976.git.u.kleine-koenig@pengutronix.de
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>

+3 -2
+3 -2
drivers/pwm/pwm-stm32.c
··· 351 351 regmap_set_bits(priv->regmap, TIM_CR1, TIM_CR1_ARPE); 352 352 353 353 /* Calculate the duty cycles */ 354 - dty = prd * duty_ns; 355 - do_div(dty, period_ns); 354 + dty = (unsigned long long)clk_get_rate(priv->clk) * duty_ns; 355 + do_div(dty, prescaler + 1); 356 + do_div(dty, NSEC_PER_SEC); 356 357 357 358 regmap_write(priv->regmap, TIM_CCR1 + 4 * ch, dty); 358 359